Biol Blood Marrow Transplant 19 (2013) 366e377

American Society for BloodASBMTand Marrow Transplantation

Fractal Organization of the Human T Cell Repertoire inHealth and after Stem Cell Transplantation

Jeremy Meier 1, Catherine Roberts 1, Kassi Avent 2, Allison Hazlett 1,Jennifer Berrie 2, Kyle Payne 2, David Hamm3, Cindy Desmarais 3,Catherine Sanders 3, Kevin T. Hogan 4, Kellie J. Archer 5, Masoud H. Manjili 2,Amir A. Toor 1,*1Bone Marrow Transplant Program, Department of Internal Medicine, Virginia Commonwealth University, Richmond, Virginia2Department of Microbiology and Immunology, Virginia Commonwealth University, Richmond, Virginia3Adaptive Biotechnologies, Seattle, Washington4Massey Cancer Center, Richmond, Virginia5Department of Biostatistics, Virginia Commonwealth University, Richmond, Virginia

Article history:

Received 26 August 2012Accepted 12 December 2012

Key Words:T cell receptorsStem cell transplantationSelf-similarityFractal

Financial disclosure: See Acknowl* Correspondence and reprint re

Center, Virginia Commonwealth U980157, Richmond, VA 23298.

E-mail address: atoor@vcu.edu

1083-8791/$ e see front matter �http://dx.doi.org/10.1016/j.bbmt.20

a b s t r a c tT cell repertoire diversity is generated in part by recombination of variable (V), diversity (D), and joining (J)segments in the T cell receptor b (TCR) locus. T cell clonal frequency distribution determined by high-throughput sequencing of TCR b in 10 stem cell transplantation (SCT) donors revealed a fractal, self-similarfrequency distribution of unique TCR bearing clones with respect to V, D, and J segment usage in the T cellrepertoire of these individuals. Further, ranking of T cell clones by frequency of gene segment usage in theobserved sequences revealed an ordered distribution of dominant clones conforming to a power law, witha fractal dimension of 1.6 and 1.8 in TCR b DJ and VDJ containing clones in healthy stem cell donors. This self-similar distribution was perturbed in the recipients after SCT, with patients demonstrating a lower level ofcomplexity in their TCR repertoire at day 100 followed by a modest improvement by 1 year post-SCT. A largeshift was observed in the frequency distribution of the dominant T cell clones compared to the donor, withfewer than one third of the VDJ-containing clones shared in the top 4 ranks. In conclusion, the normal T cellrepertoire is highly ordered with a TCR gene segment usage that results in a fractal self-similar motif ofpattern repetition across levels of organization. Fractal analysis of high-throughput TCR b sequencing dataprovides a comprehensive measure of immune reconstitution after SCT.

� 2013 American Society for Blood and Marrow Transplantation.

INTRODUCTIONT cells are central to the normal execution of adaptive

immunity, allowing identification of the multitude of path-ogens encountered in an organism’s lifetime. Immunerecognition of transformed cells further contributes tosurvival of the host organism by preventing emergence ofmalignancy. It is no surprise, then, that adoptive immuno-therapy by means of allogeneic stem cell transplantation(SCT) has emerged as an effective modality for the manage-ment of hematopoietic malignancies. Survival after SCT iscritically dependent on immune reconstitution because ofthe role of adoptive immunity in graft-versus-host (GVH)and graft-versus-leukemia (GVL) responses as well as itsobvious importance in controlling opportunistic infections.Several measures to evaluate immune recovery after SCT,such as T cell chimerism, and T and NK cell subset recovery,are correlated with post-transplant outcomes. However,none of these provides a comprehensive picture of T cellreceptor repertoire reconstitution, knowledge of which iscritical to allow full comprehension of GVH and GVLresponses, which are driven by minor histocompatibilityantigen differences between donors and recipients.

edgments on page 376.quests: Amir A. Toor, MD, Massey Cancerniversity, 1300 E Marshall St, mail stop

(A.A. Toor).

2013 American Society for Blood and Marrow12.12.004

T cell receptors (TCRs) are expressed on the T cell surfaceserving primary antigen recognition function in adaptiveimmune responses. TCRs are comprised of an alpha anda beta chain (TCR ab) each consisting of an antigen binding(complementarity determining region; CDR1-3) and trans-membrane domain. The CDR3 region of the TCR interactswith oligopeptides presented in the antigen-binding grooveof the human leukocyte antigen (HLA) molecules expressedon the antigen-presenting cells. The ability of the human Tcell repertoire to recognize the vast array of pathogens andinitiate specific adaptive immune responses depends on theversatility of the TCR, which is generated by recombinationof diversity (D), joining (J), and variable (V) segments withinthe TCR gene locus. The germ line TCR b locus has 2 D, 13 J,and 52 V gene segments, which are recombined during T celldevelopment to yield numerous VDJ recombined T cellclones. Further variability and antigen recognition capacity isintroduced by nucleotide insertion (NI) in the recombinedTCR a and b VDJ sequences. This generates a vast T cellrepertoire, the relative quantitative organization of which isthus far poorly understood. High-throughput TCRsequencing allows in-depth molecular analysis of T cellclones to get an unprecedented level of detail when exam-ining the T cell repertoire of individuals. However, tocomprehend the significance of the vast array of T cell clonesidentified and their relative quantitative relationship in thenormal and disease states, a better understanding of thenormal clonal frequency distribution of the T cell repertoireis needed.

Transplantation.

J. Meier et al. / Biol Blood Marrow Transplant 19 (2013) 366e377 367

Self-similarity is a phenomenon observed frequently inbiologic systems, and results in scale invariant motifs overa finite range in living organisms [1-3]. Readily discernibleexamples of this self-similarity include the arborizing of treebranches and vascular networks, where first order branchesare quantitatively (say, in terms of cumulative diameter)similar to second order branches and so on. Although neveridentical, such patterns are termed statistically self-similar.Self-similarity in complex natural systems is characterizedby quantitative relationships in which the magnitude of thevariable of interest varies with the scale of measurement,such that, as the scale of measurement gets smaller themagnitude becomes relatively larger maintaining constantproportionality. This relationship is expressed in terms offractal dimension, which takes on a noninteger valuebetween the classical Euclidean integer dimensional values,and is given by the proportion of the logarithm of magnitudeby logarithm of the scale [4,5]. The transformation of naturalnumbers to their logarithms allows the proportionality ofmagnitudes across different scales to be more easily dis-cerned. Fractal dimension, then, is a measure, which explainsthe complex structural organization of natural objects, whichdo not completely occupy the space they exist in. As anexample, in Euclidean geometry, a plane occupies the spacecompletely in 2 dimensions, and a cube in 3. However innature, structures do not completely occupy the space theyexist in when viewed from a topographical point of view. Inthe above examples, vascular networks and tree branches donot completely occupy the 3-dimensional space they exist in;they, irregularly, occupy less. Thus, they have a fractaldimension of between 2 and 3. Similarly a linear object, suchas a coastline or a meandering river, will have a fractaldimension of between 1 and 2, because it is a one-dimensional object, which is zigzagging in and out ofa second dimension on account of structural complexityintroduced by folding.

T cell receptor repertoire generated by recombination ofgene segments arranged on a linear DNA molecule maylikewise have a fractal organization, with the clones con-taining different V, D, and J gene segments having a propor-tional quantitative distribution in an individual. Wepostulated that the TCR bD, J and V gene segment usage in anindividual would result in a TCR repertoirewith a self-similarfrequency distribution of T cell clones incorporating thesegene segments (Figure S1). To determine this, the T cellrepertoire of donors and recipients of HLA matched relatedand unrelated allogeneic SCT assessed by high-throughputsequencing of the TCR b was analyzed using log-transformation and fractal analysis.

MATERIALS AND METHODSPatients

Patients with recurrent hematological malignancies were enrolled ina prospective clinical trial approved by the Virginia CommonwealthUniversity Institutional Review Board (Clinicaltrials.gov identifier:NCT00709592). The study is a randomized phase II trial of a reducedintensity conditioning regimen for allogeneic SCT in patients with recurrenthematological malignancies. High-resolution HLA matching was performedat HLA-A, B, C and DRB1 loci in matched related as well as matched unre-lated donors, with 7 of 8 or 8 of 8matching required for transplant eligibility.Patients were randomly assigned to conditioning with 2 different doses ofrabbit antithymocyte globulin (Thymoglobulin, Genzyme, Inc., Cambridge,MA) and 450 cGy total body irradiation, followed by infusion of granulocyte-colony stimulting factor mobilized blood stem cells as previously described[6]. Chimerism analysis was performed on CD3þ T cells by polymerase chainreaction (PCR) for STR sequences to identify informative donor-recipient locifor measuring recipient contribution to the CD3þ cell population incirculation.

T cell Isolation, RNA Extraction and RT ReactionRNA was extracted from donor apheresis product and recipient CD3þ

PBMC using TRIzol reagent (Invitrogen) according to manufacturer’sprotocol. DNase treatment was performed using RQ1 RNase-free DNase(Promega, according to manufacturer’s protocol) to digest DNA contami-nation. The cDNAwas prepared from 1 mg of total RNA using the SuperScriptII reverse transcriptase (Invitrogen) with a dT18 oligonucleotide primer.cDNA synthesis was completed at 42�C for 2 hours [7]. Recipient bloodsamples were collected at day 100 and one year after SCT in all patients.Patients developing graft-versus-host disease (GVHD) had samples collectedbefore starting corticosteroids, whenever possible.

High-Throughput T Cell Receptor SequencingUpon confirmation of the purity of the cDNA by running PCR product of

GAPDH amplification,1 mg to 119 mg (average, 55 mg) per sample of cDNAwassent to Adaptive Biotechnologies (Seattle, WA) for high-throughputsequencing of the TCR b CDR3 region using the ImmunoSEQ assay. Thisapproach is composed of a multiplex PCR and sequencing assay in combi-nation with algorithmic methods to produce approximately 1,000,000 TCRb CDR3 sequences per sample [8]. The assay utilizes 52 forward primers forthe Vb gene segment and 13 reverse primers for the Jb segment to generatea 60 base pair fragment capable of identifying the entire unique VDJcombination [9]. It is to be noted that primer bias is inherent to multiplexPCR, but is predictable and reproducible at all V, D, and J gene loci that can beamplified; thus, this assay utilized both experimental and computationaltechniques to normalize for bias and produce a quantitative assay. Ampli-cons were then sequenced using the Illumina HiSeq platform, and data wasanalyzed using the ImmunoSEQ analyzer set of tools. This approach enablesdirect sequencing of a significant fraction of the TCR repertoire as well aspermitting estimation of the relative frequency of each CDR3 region in thepopulation.

Proportional Usage of TCR b Gene SegmentTo study the quantitative organization of the T cell repertoire with

respect to the VDJ gene segment usage, productive sequences were exam-ined after high-throughput sequencing. The TCR gene segments werechosen as a natural scale for examining the organization of the TCR reper-toire. Clone frequency (copy [or read] number of individual, uniquesequences bearing specific TCR b J, V, and NI) was used as a measure ofmagnitude and the number of segments incorporated was used as a scalingfactor. Scaling factors (SF) were assigned to sequences as follows: sequenceswith the same TCR b J segment were assigned a SF of 2, because 2 genesegments are involved (D and J); those with TCR b VJ were assigned a SF of 3(D, J and V involved); and unique sequences bearing nucleotide substitu-tions were assigned a SF of 4 (D, J, V, and NI involved) (Figure S1). Clonefrequency of specific J segment usagewas the sum of sequences bearing thatJ segment, and clone frequency was calculated similarly for the specific VJclones, whereas clone frequency of individual NI containing sequences wasas reported, based on that sequence’s copy number.

High-throughput TCR sequencing gives clone frequency data with geneusage information at the greatest resolution, analogous to measurement atthe smallest scale. In this system, the Tcell clone frequency becomes smalleras the scale of gene segment usage assumes higher values, and thus reso-lution (from J, to VJ segment down to VJþNI). This configuration is reciprocalto other fractal structures in which usually the magnitude proportionallyincreases as the scale of measurement becomes smaller (ie, magnificationincreases). Therefore, analogous to the fractal dimension calculation, inorder to demonstrate fractal distribution of TCR b gene segments acrosssamples, we multiplied the log of the number of sequences containing thegene segment being measured, by the log of scale, ie, the level of genesegment being considered. In the absence of unique data sets at each of theTCR b gene segments, this method accounted for the entire data set beingmeasured at the highest scale, and then split up by gene segment usage.

The formula used was:Let Xijkl represent the clone frequency for patient, i at time j, for the kth J,

and lth V segment. Donor samples are denoted by j¼1 and recipient samplesare j> 1. For subject i, at time j, our self-similarity score from J segments wasestimated as, (formula 1)

SSJij ¼ meank

log

X47l¼1

Xijkl

!� logð2Þ

!Formula 1

and from VJ segments as, (formula 2)

SSVij ¼ meank

�mean

l

�log�Xijkl

�� logð3Þ

��Formula 2

Using m to denote nucleotide insertions, our self-similarity score fornucleotide insertions were calculated as, (formula 3)

J. Meier et al. / Biol Blood Marrow Transplant 19 (2013) 366e377368

SSNIij ¼ meanm

�log�Xijklm

�� logð4Þ

�: Formula 3

Therefore, the general form of our self-similarity score was,

SSTCR ¼ meanðlogðXiÞ � logðSFxÞÞ

where, Xi ¼ TCR b clone frequency (unique sequence copy number) withunique J, VJ, or VJþNI, and corresponding SFx¼ 2 for J, 3 for VJ and 4 for VJþNI.

For these calculations, any undefined segment generated during thesequencing steps, where the TCR b V family could not be assigned, wasexcluded. The self-similarity score for NI used sequences present at greaterthan 100 for the copy number after normalization for PCR primer bias. Thiswas done to capture the organization of the dominant contributors to therepertoire in order to discern the large-scale organization.

Graphical Representation of TCR b RepertoireSequencing data for the TCR b repertoire generated by Adaptive

Biotechnologies was organized according to VDJ clonotype. Individual graphswere generated using Microsoft Excel’s doughnut graph function, whichallowsfor thedepictionof relative-proportional distribution (RPD)ofmultiplevariables simultaneously. Each circle within the graph represents 100%, andhence, each segment can be visualized based on its proportional represen-tationwithin thepopulationbeing inspected.Briefly, foreachTCRbDsegmentgraph, each ring represents a different TCR b J segment, where, as you movefurther from the center of the graph, with each ring you move further alongthe TCR b J gene segment (J1.1/J2.7). Each TCR b J segment graph representstheproportion towhich J1.1/J2.7 is expressed in each sample. Similarly, eachTCR bV RPD graph depicts a different V segment for each ring moving fromV2/V30,with eachV segment brokenupbased on the frequencywithwhichit is found spliced with each J segment. There are 47 rings in each V segmentgraph, as the primers for the TCR b V segment used by ImmunoSeq candifferentiate between 48 different V segments. The primer set could notdifferentiate TRBV12-2 and TRBV12-3 from each other; nor TRBV6-3 andTRBV6-4. Notably, TRBV17 was excluded from most analyses (self-similarityscores and RPD graphs) because of absent expression in most cases for bothdonor and recipients. Lastly, nucleotide insertions were displayed using RPDgraphs such that the total number of unique nucleotide insertion bearingclones for both the Vb-Db and Db-Jb were represented. Each ring in the graphrepresents a different TCR b J segment (J1.1/J2.7) that is broken up propor-tionally based on the number of sequenceswith unique nucleotide insertionspresent at each variable region, expressed as a fraction of all the sequencescontaining that J segment. Results from TCR bD1 and D2 segment plots werecombined in the represented graph because of their similarity.

TCR bClonal Frequency DistributionSelf-similarity describes scale invariance of magnitude, and is classically

modeled by a power-lawdistribution, such that a function’s value, y, varies asa constant power, a, of its argument, x, maintaining a proportional rela-tionship between the two variables, resulting in the relationship, y ¼ k xa.Among other physical processes, such power law distribution describesallometric scaling invariance in biologic systems [10-13]. The self-similar,and thus fractal, power law distribution of high-throughput TCR clonalfrequency was first studied by constructing frequency distribution plots ofthe TCR bDJ, VDJ and VDJþNI clones. Subsequently this relationship wasverified by creating log-log plots of assigned rank against the relativefrequency of those clones, with rank assigned by relative frequency. A linearrelationship between these base 10 log-transformed variables is describedby the formula log y¼ a log xþ log k, [formula 4] where the absolute value ofthe slope of the line, ‘a’, is considered equivalent to the fractal dimension. Forthis calculation, relative clonal frequency was determined by taking the sumof the individual clone frequencies (see above) at the segment level to bestudied, ie, at either DJ and VDJ, and then dividing the clone frequency ofeach unique, DJ and VDJ clone being evaluated, by the total clonal frequencyat that level to determine the relative usage of individual clones. Sequenceswith fewer than 500 copies were excluded from this analysis. Rank assign-ment for VDJ clones was based on the prevalence of those clones whosefrequency in the overall pool was greater than .05%, to determine theorganization for the dominant T cell clones in the repertoire. The rankingbased onpercent frequency (relative frequency� 100, f) was done as follows:

RVDJ ¼

8>>>>>>>>>>>><>>>>>>>>>>>>:

1 if f � 1%2 if :50% � f < 1%3 if :25% � f < :50%4 if :15% � f < :25%5 if :10% � f < :15%6 if :075% � f < :10%7 if :06% � f < :075%8 if :05% � f < :06%NA otherwise

Formula 4

The difference between the consecutive ranks was consistent with thefollowing progression between ranks: .5, .25, .1, .05, .025, and .01. For DJclone assignment in the log-rank plots, the segments D1J1.1 to D2J2.7 wereranked as follows:

RDJ ¼

8>>>>>><>>>>>>:

1 if f � 7%2 if 4% � f < 7%3 if 2% � f < 4%4 if 1% � f < 2%5 if 0:4% � f < 2%NA otherwise

Segments that were represented at a frequency of less than .4% were notincluded for the DJ rank plots. The difference between the consecutive rankswas consistent with the following progression between ranks: 3, 2, 1, and .5.The absolute value of the slope of the resulting linear regression lines fromthe points was used as a measure of the self-similarity and fractal dimensionof the donors.

To determine changes in TCR repertoire after SCT, we examined the 4dominant ranks in the donors and recipients, and reported the proportion ofVDJ containing clones, which were present in both donors and recipients inthese top ranks. These would be most likely harbor the dominant sequences(w200 in number), and would arguably include the T cell clones drivingGVHD and response to infection. Sequence reads where the variablesegment was undefined were generally excluded in this analysis.

Statistical AnalysisStatistical analysis was performed with SPSS (version 20) software. The

Pearson Correlation Coefficient was used to study the relationship betweenthe power law of the TCR b DJ and VDJ clones. Wilcoxon signed rank test wasused to calculate the difference between the self-similarity scores for paireddonor-recipient samples at day 100 and 1 year post-transplant. Statisticaltests were based on a level of significance of P < .05.

RESULTSSelf-Similarity of the TCR b Repertoire

High-throughput sequencing of the TCR b locus wasperformed using cDNA isolated from either apheresisproduct (stem cell donors) or circulating CD3þ cells (SCTrecipients) at day 100 and one-year post-SCT or at the time ofdiagnosis of GVHD (n ¼ 10, Table S1). A median 629,606productive unique TCR b sequences were identified (range:1,161,823 to 58,009) in the stem cell donors. TCR b clonefrequencies were then examined to determine the TCR self-similarity score (SSS), evaluating clone frequency at thelevel of each gene segment (J, VJ, and VJþNI). The J segmentusage in individuals was first examined as it recombines withthe D segment early in T cell development. SSS for each TCRb J gene segment (J1.1 to J2.7) was found to take on a limitedrange of values across individual donors as depicted(Figure 1A). This finding of relatively similar gene segmentusage was consistent across all rearranged gene segmentlevels (scales) and between all the donors (Table 1). Thesevalues were limited to a narrow distribution between 1.4 and1.6, and demonstrated a trend of declining values as thescaling factor (gene segment usage) increased. This obser-vation is consistent with the notion that when measuringfrom the J segment, up, the clones become more preciselydefined, albeit with a greater number of gene segmentsinvolved in clonal definition. Further, the observed closeproximity of the SSS of TCR b J, VJ and VJþNI usage suggeststhat there is an underlying organization to the TCR reper-toire, and that the gene segment usage observed here is notrandom but rather, similarly distributed across the TCRb gene segments and amongst individuals.

Further evidence for self-similarity across the T cellrepertoirewas demonstrated by the RPD graphs, which allowvisual estimation of the relative proportion of TCR genesegment usage both within and across donor samples bydepicting relative clonal frequency. When looking at theproportional representation for the TCR b J segment using

Figure 1. Self-similarity in TCRb J gene segment usage across segments andindividuals. (A) Self-similarity score for each TCR bJ gene segment (J 1.1 to J2.7), depicting the narrow range of distribution for each J segment in all thedonors. (B) Graphical representation of self-similarity. Relative-proportional-distribution (RPD) graph depicting TCR b J segment usage in 10 hematopoieticstem cell donors. Each ring in the graph represents J segment usage frequencyin a single donor, showing similarity in J segment frequency across differentdonors. Arrows demonstrate data orientation in the rings.

Table 1Self-Similarity Score of the TCRbClone Frequency (mean � SD) at DifferentGene Segment Levels (X). T Cell Receptorb SSS ¼ Log (Xi) � Log (SFx)

Donor SSS-TCR-J* SSS-TCRb-Vy SSS-TCRb-NIz

3 1.78 � .12 1.74 � .20 1.42 � .214 1.58 � .14 1.45 � .19 1.38 � .175 1.65 � .07 1.50 � .12 1.41 � .206 1.82 � .07 1.76 � .10 1.43 � .217 1.65 � .07 1.49 � .12 1.42 � .208 1.70 � .06 1.59 � .10 1.41 � .199 1.71 � .13 1.65 � .20 1.36 � .16

11 1.41 � .11 1.21 � .15 1.41 � .2114 1.75 � .09 1.64 � .14 1.47 � .2516 1.50 � .10 1.28 � .13 1.39 � .19Average 1.65 � .13 1.53 � .18 1.41 � .03

Formulas 1, 2 and 3 respectively (See Methods, values depicted rounded offto the nearest hundredth).

* mean of n ¼ 13.y n ¼ 594.z n ¼ 6681, clones containing relevant TCRb.

J. Meier et al. / Biol Blood Marrow Transplant 19 (2013) 366e377 369

this graphic there is marked relatedness across donors(Figure 1B). The distribution of TCR bDJ sequences within thedonor pool was quantified next, and a consistent patternwasobserved within and between the healthy donors, witha preference towards the D1 splice product over D2 (Figur-es 2A, S2A). Moving further within the TCR b gene locus thispatterning was evident for the variable segments as well(Figures 2B, S2B). Similar organization was evident even atthe nucleotide insertion level when the proportion of TCRwith unique sequences, harboring different V, D, and Jsegments are evaluated (Figure 2C).

T Cell Clone Frequency DistributionEach RPD graph, when examined on its own, reveals self-

similarity across T cell clones, particularly at the VJ and NIlevels (Figure 2B-C). When viewing the circular graph ina clockwise fashion the distribution of dominant andminor Tcell clones is essentially indistinguishable within an

individual repertoire. Thus, if the RPD graphs were to bedivided into relatively equal segments, T cell clonal distri-bution appears symmetrical with different segments con-taining a few highly expressed clonotypes scattered withmultiple low-frequency ones. To mathematically verify thisTCR clonal hierarchy, clone frequency distribution and rank-frequency relationships of the clones (incorporating differentV, D, and J segments) within the T cell repertoire wereinvestigated. Nonlog transformed TCR clone frequency whenplotted in descending order, followed a power distribution,y ¼ k xa, demonstrating an asymptotic decline with a limitednumber of high-frequency clones and a large number ofclones present at a low frequency (Figure 3). The greater thedepth of sequence identity (TCR b DJ versus VDJ versusVDJþNI defined clones) the better defined the power lawrelationship became (Table 2). Significantly, the TCR Vb clonal frequency distribution curve (Figure 3) exponent, a ,took on a uniform set of values in the dominant T cell clonesexamined (Supplementary Table 2), supporting the notion ofself-similarity of the overall T cell repertoire between indi-viduals under steady state conditions. To explore the rela-tionship between a and the logarithmically derived SSS foreach donor, the base of natural logarithms, e (2.7818), wasraised to the power of a to obtain the numeric value forwhich it is the natural logarithm. The product of this numberand the constant p (3.1415) yielded a function of a, pea whichclosely approximated the SSS calculated earlier at the VJ(Kruskal-Wallis Test P ¼ .15) and VJþNI (P ¼ .288) scalinglevel, with greater congruence between the 2 valuesobserved as Tcell clonal definition increased (SupplementaryTable 2). Thus, this TCR clonal frequency distribution func-tion (pea) independently validates the SSS observed in earliercalculations and identifies the mathematical constraintswithin which the T cell repertoire appears to be organized.

For the log rank-frequency analysis, the relative frequencyof each clone within the repertoire was determined and theforemost ranks were assigned to the most highly expressedTCR b DJ or VDJ clonotypes within each donor, with subse-quent ranks being used to describe less abundant clones. Aconsistent trend was observed in which a small number ofdominant, highly expressed T cell clones constituteda majority of the repertoire occupying the top ranks anda large number of T cell clones occupied the later ranks(Supplementary Table 3A and 3B). The resulting relative TCRclonal frequency, when plotted against rank on a log-log plot,

Figure 2. Graphical representation of self-similarity in stem cell donor T cells. (A) Self-similarity in D segment usage, each ring in the RPD graph represents one Jsegment. (B) Self-similarity and proportional distribution of VJ segment usage. Each ring represents a unique V segment with J segment usage in that V segmentdepicted by colors. (C) Nucleotide insertions across all junctions, each ring depicts one J segment, with number of unique sequences (bearing unique nucleotideinsertions) at each V locus expressed as a proportion of total number of unique sequences across that J segment. D1 and D2 are combined into one graph. Arrowsdemonstrate data orientation in the rings.

J. Meier et al. / Biol Blood Marrow Transplant 19 (2013) 366e377370

yielded a straight line for the frequently expressed clones(those with � .4% contribution to the repertoire for DJ, and �.05% for VDJ), once again, consistent with the power lawdistribution for TCR clonal frequency within the limits beingexamined, ie, in the dominant clones (Figure 4). These plotswere comparable among donors when looking at both TCRb DJ, and VDJ clones with the resulting absolute value of theslopes averaging 1.6 and 1.8 respectively (Table 3). From thisit can be inferred that a clonal hierarchy exists in each indi-vidual, where there are certain dominant T cell clones andother minor clones, which together have a self-similardistribution across the T cell repertoire when analyzed byTCR gene segment usage. An analogy will be the size distri-bution of stones in a riverbed, where individual stones havedifferent sizes but the distribution of stones of different sizesis similar from one area of the river bed to the next. Theseresults support the notion that TCR b gene segment usageacross individuals has a fractal organization.

TCR Self-Similarity with Comparable HLA TypeA potential source for the orderliness (and inter-

individual variation) that is observed in the TCR b poolwithin individuals may originate within the human leuko-cyte antigen (HLA) repertoire of each individual. If this is the

case, individuals with comparable HLA typing should havea corresponding increase in their overall relatedness at thelevel of TCR b expression. Two donors (3 and 9) werepartially HLA matched and shared an HLA haplotype,although none of the others did (Table 4, SupplementaryTable 4). When the TCR bJ segment distribution for donor 3is compared to the partially HLA-matched donor 9, versusHLA disparate donor 4, it is apparent that the frequency withwhich segment usage is seen between donors 3 and 9 isalmost indistinguishable (Figure 5A). Although there is stilla similarity in the representation between donors 3 and 4(Figure 5A), there is not the same extent of overlap as thatseen between donors 3 and 9, who are more analogous at theHLA level. Although more subtle, because of complexity, andvariables such as a single haplotype being matched andpresumably vastly different lifetime antigen exposure inunrelated individuals, this observation is also true for thediversity and variable segments, as well as at the nucleotideinsertion level (Figure 5B).

TCR b Clonal Shift in SCT RecipientsThe make-up of the TCR b repertoire in SCT recipients

from these donors was studied to explore whether or notthey retained self-similar characteristics. The self-similarity

Figure 3. Self-similar distribution of TCRb clone frequency depicted asa power function (y ¼ k xa) when clone frequencies are plotted in descendingorder, with respective R2. (Top Panel) TCRb DJ clones; (Middle Panel) TCRb VDJclones; (Bottom Panel) TCRb VDJþNI clones.

Table 3Self-Similarity in Frequency Distribution of T Cell Clones. Absolute Value ofthe Slope of the Log-Log Plot of Frequency-Ranked TCRbDJ and VDJ Clones inEach Stem Cell Donor (a in log y ¼ a log x þ log k)

Donor DJ Rank Slope R VDJ Rank Slope R

3 1.65 .95 1.76 .984 1.66 .95 1.84 .985 1.73 .92 1.81 .986 1.49 .92 1.81 .977 1.41 .92 1.77 .988 1.49 .95 1.78 .979 1.70 .93 1.78 .98

11 1.82 .93 1.84 .9614 1.41 .91 1.80 .9816 1.62 .91 1.90 .98

Clones with sequence counts < 500 were excluded. All correlations signif-icant at < .0001 (Pearson’s correlation coefficient).Formula 4 (see Methods).

J. Meier et al. / Biol Blood Marrow Transplant 19 (2013) 366e377 371

score was applied to high-throughput sequencing data fromtransplant recipients at day 100 post-transplant, and again at1 year with available data. A reduction in the self-similarityscore is apparent for recipients at day 100 for both the TCRb J and V segments with average values of 1.3 and 1.1respectively, almost always declining compared to the donor(P ¼ .047 for J, and P ¼ .022 for V) (Tables 5 and 6, Figure 6A-B). However, both the self-similarity score for Jb and Vb

increased to levels comparable to that seen in donors withaverage values of 1.7 and 1.5 at 1 year (P ¼ .893 for J and

Table 2Correlation Coefficient, R, for Power Law Distribution of TCR b DJ, VDJ, andVDJþNI clones

Donor TCRb DJ* TCRb VDJy TCRb VDJþNIz

3 .86 .99 .994 .87 .99 .995 .83 .99 .996 .86 .99 .997 .81 .99 .999 .88 .99 .998 .85 .99 .99

11 .92 .99 .9914 .84 .99 .9916 .86 .99 .99

P value for all correlation coefficients was < .0001.* n ¼ 26 clones.y n ¼ 1292 � 995 clones.z n ¼ 6681 � 4923 clones.

P ¼ .345 for V) (Tables 5 and 6). These deviations from thedonor repertoire are more apparent when looking at therelative proportion of variable and joining segments visually.

RPD graphs from SCT recipients from donor 4 and donor 6were utilized to look more closely at the TCR b pool at bothday 100 and 1 year post-transplant. The complexity of order,and similarity that had been apparent within the apheresisproducts from the donors is hardly perceptible in bloodsamples obtained from SCT recipients, especially earlier onpost-transplant (Figure 6C). This decline in TCR complexitywas evident throughout the recipient population at day 100(Figure S3). Despite the lag in complexity of the T cell orga-nization in recipients at 100 days post-SCT, the recipient Tcell repertoire appears to be trending towards a moreordered and complex arrangement like that seen in thedonors at 1 year (Figures 6C, S3).

Using the TCR clonal frequency ranking analysis, thedominant Tcell clones, that is, those represented in the first 4ranks, were compared between donors and recipients at day100 and 1 year after SCT. When comparing these dominantclones between donor and recipient at day 100 and 1 year,patients shared only a small proportion of TCR clonotypeswith their donors independent of full donor chimerism(Table 7). Upon comparing patients who had GVHD withGVHD-free patients, there was a weak association observedbetween the extent of disparate TCR clones and GVHD inci-dence. The difference in the degree of overlap observed inthe top 4 ranks between donors and recipients at day 100averaged 22% versus 32% clones shared between the 2 groupsrespectively (Pearson correlation �.78, P ¼ .03) (Table 7). Thenegative correlation implied that development of GVHD iscorrelated with declining number of shared clones. Thoughthis difference did not change over time, its significance waslost with declining patient numbers. In aggregate, thesefindings suggest an alternate T cell clonal hierarchy andrepertoire develops in transplant recipients when comparedwith their donors.

DISCUSSIONHigh-throughput sequencing of TCR b families demon-

strates that the organization of the TCR b repertoire, asdetermined by TCR gene segment usage in normal stem celldonors, rather than being random, is highly organized ina self-similar pattern with a hierarchy of dominant andminor clones. Relative proportional TCR gene segment usageand the resulting repertoire may be determined by the HLAmake up of an individual. Further the complexity of the TCR

Figure 4. Log-Log plot depicting linear distribution of relative clone frequency ranked TCRb DJ (Top Panels) and VDJ (Bottom Panels) clones in two stem cell donors.

J. Meier et al. / Biol Blood Marrow Transplant 19 (2013) 366e377372

repertoire is diminished after SCT, and that TCR b hierarchy isaltered possibly contributing to the development of immu-nologic sequela such as GVHD.

Self-similarity at different scales is widely observed innature and is generally associated with physical structure,where measurements across scales are possible. It is notintuitive that such self-similar motifs with a fractal order willbe evident at a molecular level in a system bereft of physicalstructure. However, the data presented support the notionthat the T cell repertoire, which is generated by recombina-tion among multiple gene segments located at the TCR locus,results in fractal ordering with respect to gene segmentusage, which is evident in the subsequent clonal hierarchyobserved. High-throughput sequencing studies have startedto unveil this organization, as in a recent study examiningrepertoires of naïve and memory CD8þ T cells, whichdemonstrated a restricted and overlapping repertoire inseven individuals [14]. The Db loci were proportionally andconsistently rearranged with the gene segments in the 2 Jbclusters. Non-uniform VbJb usage was compared with theexpected probabilities in a model in which all the NI possi-bilities are accounted. However, within the observed reper-toire, more than 10,000 sequences were shared betweenthese 7 individuals. The authors hypothesized that therepertoire is dependent on the antigenic peptides encoun-tered by the individual. The data presented here confirm andextend the notion of repertoire restriction and ordering byrevealing the self-similar distribution, which results ina hierarchical T cell clonal distribution.

Table 4HLATyping Results for Three Donors: Twowith Partial HLAMatching (Donor 3 and DVersus Donors 3 and 9)

Donor HLA-A HLA-A HLA-B

3 03:01 11:01 07:029 03:01 07:024 23:01 30:02 15:03Donor DQB1 DQB1 DRB13 05:03 06:02 14:01/549 05:01 06:02 01:034 02:01/2 03:01

Asymmetric use of TCR gene segments in generating the Tcell repertoire has been reported when high-throughputsequencing was applied to a TCR RNA pool from severaldifferent individuals [15]. Both TCR b V and J segmentfrequency distribution plots observed were reminiscent ofa power law distribution with abundant (V 20-1, 5-1,29-1.and J 2-1,1-1, 2-7.) and infrequent clones (V 5-3, 5-5,5-6.and J 1-3, 2-4, 2-6.). Rank-frequency relationshipshave previously been reported as a means of establishing notonly the self-similar clonotype distributionwithin a subset ofthe T cell repertoire, but also as a method to describe thefractal nature of this population [16]. In an elegant series ofpapers examining observed frequencies of specific clono-types after viral infection, a power law relationship wasevident [16,17]. Specifically, clonotypic T cell responses toinfluenza peptide MI58-66 presented on HLA-A2 character-ized by TCR BV17, had a biphasic fractal distribution witha larger number of infrequent clones and a small number ofhighly expressed clones, with the former rapidly decayingtowards the latter, and when these clones were rankedaccording to frequency, a power law relationship wasobserved. Similar deterministic organization has beenobserved in the zebrafish immunoglobulin VDJ repertoire,where the repertoire is demonstrably comparable among theyoung fish [18,19]. However, the VDJ repertoire diversifies inolder animals by eliminating the effect of clonal amplifica-tion. These authors discovered that independent of age, theunderlying “primary” repertoire was maintained througha deterministic program rather than a stochastic process

onor 9) and Onewith Complete HLAMismatch with the Other Two (Donor 4

HLA-B HLA-Cw HLA-Cw

55:01 03:03 07:0207:02

44:03 02:10 07:18DRB115:0115:0107:01

Figure 5. Resemblance of T cell repertoire in donors with partial HLA match, and lack thereof in HLA mismatched donor. (A) J segment usage. (B) DJ, VJ and nucleotidesubstitution.

J. Meier et al. / Biol Blood Marrow Transplant 19 (2013) 366e377 373

determining the immunoglobulin repertoire. These findingsare supported by our frequency distribution analysis of high-throughput TCR b sequencing data.

In determining the organization of the TCR bVDJ reper-toire by gene segment usage, regardless of the method used,we almost always arrived at a self-similarity score or fractaldimension value between 1 and 2. This is consistent with thenotion that splicing and recombination of linear germ-lineDNA constituting the TCR locus, results in a virtual, non-

Euclidean, higher dimensional configuration. Indeed, ourFD and SSS are strikingly close to the value of 1.6 previouslyreported [16]. Logically, germ line TCR locus DNA will havea dimensional value of 1, and recombination will yielda higher dimensional value, similar to other linear motifs innature, such as coastlines. Interestingly, this value declines asthe number of gene segments utilized increases, consistentwith greater precision as one examines the data set at greaterresolution from DJ to VDJ to VDJþNI substituted clones

Table 5TCR b J and VJ SSS in SCT recipients at day 100 and 1 year Formulas 1 and 2(see Methods)

Recipient SSS-TCR b JDay 100*

SSS-TCR b J1 Year*

SSS-TCR b VJDay 100y

SSS-TCR b VJ1 Yearz

3 1.62 � .17 NA 1.34 � .22 NA4 1.59 � .08 1.68 � .13 1.38 � .11 1.54 � .185 1.62 � .12 NA 1.43 � .17 NA6 1.65 � .13 1.69 � .11 1.40 � .14 1.50 � .137 .84 � .35 1.74 � .09 .67 � .20 1.56 � .148 .63 � .38 1.74 � .12 .51 � .23 1.54 � .139 1.58 � .15 1.63 � .11 1.39 � .20 1.46 � .14

11 1.64 � .12 NA 1.41 � .14 NA14 1.65 � .10 NA 1.49 � .13 NA16 .62 � .15 NA .46 � .10 NAAverage 1.34 � .45 1.70 � .05 1.15 � .42 1.52 � .04

* n ¼ 13.y n ¼ 412.z n ¼ 562 clones containing relevant gene.

J. Meier et al. / Biol Blood Marrow Transplant 19 (2013) 366e377374

(Figure S1). The mathematical constraints governing theT cell repertoire organization also were evident in therelationship observed between the SSS calculated from log-transformed data and the T cell clonal frequency distribu-tion function (pe a )obtained from non-transformed numericdata. It is intuitive that constants of nature such as e and pshould be observed when recombanitorial possibilities areconsidered in double helical DNA. An obvious source ofvariability in the calculated SSS and FD is the differentnumber of recombinatorial possibilities at each genesegment, and there is evidence that V-J recombination biasgenerating the naïve repertoire may be the primary biolog-ical force responsible for shaping the TCR repertoire [20].Such nonuniformity due to specific mechanisms is entirelyconsistent with the paradigm that fractal organization innature is observed on a limited scale, not extending beyondcertain orders of magnitude. The mechanism for hierarchyemerging remains to be elucidated. However, marked simi-larity in TCR gene segment usage seen in a pair of partiallyHLA-matched individuals points to HLA being a potentialcontributing factor.

Plausibly, if the HLA type of an individual significantlycontributes to the organization of the TCR repertoire, certainTCR configurations would be more efficient at identifyingantigens in the context of relevant HLA than others and beover represented. On the other hand, TCR clonotypes thatare less likely to interact with the individual’s HLA or arelikely to trigger autoreactive processes, are likely to berelegated to lower frequencies and absence. Certain low-frequency clonotypes are needed because they may be theonly ones that can interact with particular peptide-HLA

Table 6Percent Change in TCRb J and VJ SSS from Donor to Recipient at Day 100 andOne Year

Recipient Day 100TCRb-J

1 YearTCRb-J

Day 100TCRb-VJ

1 YearTCRb-VJ

3 �9.0 NA �23.0 NA4 .63 6.3 �4.8 6.25 �1.8 NA �4.7 NA6 �9.3 �7.1 �20.5 �14.87 �49.1 5.5 �55.0 4.78 �62.9 2.4 �67.9 �3.19 �7.6 �4.7 �15.8 �11.5

11 16.3 NA 16.5 NA14 �5.7 NA �9.1 NA16 �58.7 NA �64.1 NA

combinations. Low-frequency clonotypes may also accountfor naturally occurring regulatory T cell populations andexert a tolerizing influence on potential autoreactive clones[21]. Based on this analysis of high-throughput TCRsequencing data, we propose a model where TCR gene lociundergo rearrangement to generate a self-similar fractalrepertoire, in which each of the V, D, and J segments areproportionally represented. For the TCR b locus, the Dsegments form the foundation of the repertoire with a largemajority of sequences having either D1 or D2 incorporated.Each D segment recombines with the J segments (J1.1 toJ2.7) in a proportional manner, such that the 2 D segmentcontaining groups are each split in 13 different groups. Afterthis recombination, each of these (w2 � 13) groups is thenproportionally recombined with the approximately 52 Vsegments to result in multiple (w2 � 13 � 52) individualclonotypes. Nucleotide substitution at the V/D and D/J splicesites generate further diversity resulting in hundreds ofsequences for each VDJ recombination forming a completerepertoire on this, by now very diverse, 3-tiered foundation.In other words, D, J, and V segments, in that order, providea branching scaffold, with a fractal structure upon whicha “canopy” of nucleotide substituted TCR sequences isarranged (Figure 7A). The relative frequency of variousrecombinatorial possibilities is ranked hierarchically, onceagain in self-similar fashion (Figure 7B), with the dominantsequences determined by an individual’s HLA type andminor histocompatibility antigen make-up, which in turn isgoverned by the need for efficient pathogen oligopeptidepresentation and deletion of auto-reactive self-recognizingsequences. Thus, the T cell repertoire efficiently identifiespathogens from amongst all the self-antigens. This is anal-ogous to the way tree branching patterns maximize theabsorption of sunlight by the efficient usage of space.Similarly, T cell sequence ranks create a “net” without holesto safeguard the organism from extraneous harm withoutinjuring itself. Similar considerations would apply to theTCR a locus, which combined with the TCR b yields a stag-gering number of possibilities for motif recognition, on theorder of 1012. Thus, one may speculate that within eachindividual, a branching TCR repertoire yields an efficientmethod to cover the potential spectrum of pathogensencountered using the least amount of genetic information.Further, the frequency distribution, following the power lawsuggests a state of dynamic equilibrium where a smallnumber of dominant clones are in circulation at any giventime, and a large cadre of minor clones are “at the ready” inthe event of exposure to a specific pathogen.

Immune recognition of transformed cells contributes tosurvival of the host organism, preventing emergence ofmalignancy. Therefore, it is no surprise that adaptiveimmunotherapy by means of allogeneic SCT has emergedas an effective tool for the management of hematopoieticmalignancies. Survival after SCT is critically dependent onimmune reconstitution because of the role of adaptiveimmunity in GVH and GVL responses as well as itsobvious importance in controlling opportunistic infections.Several measures to evaluate immune reconstitutionincluding T cell chimerism, T cell subset recovery anddonor-derived T cell counts are correlated with post-transplant outcomes. Analysis using TCR spectratypingpost-transplant has demonstrated massive perturbation ofthe normal T cell repertoire in the transplant recipient[22]. Further, immunological complications of SCT, such asGVHD are associated with donor-derived oligoclonal T cell

Table 7TCRb Clonal Shift Post-SCT. T Cell Chimerism Data for Recipients at Day 100and One Year Post-Transplant Depicting Percentage Recipient Derived TCells

Recipient Day 100 1 Year GVHD no GVHD GVHD no GVHD

3 11 - - .39 - ND4 51 8 - .21 - .315 0 .23 - * -6 0 0 .24 - .36 -7 52 56 - .29 - .248 2 6 .13 - .24 -9 2 0 .21 - .22 -

11 0 - - .40 - ND14 2 - .31 - * -16 5 - - .32 - *

Mean .23 � .06 .32 � .08 .27 � .08 .28 � .05

The fraction of TCR VDJ clones shared between donor and recipients in thetop 4 ranks at day 100 and 1-year post-SCT depicted. GVHD status at time ofmeasurement.ND indicates not done.

* Deceased at 1 year.

Figure 6. Change in self-similarity scores over time following SCT. (A) Change in TCRb J SSS from donor to recipient at Day 100 and 1 year post-transplant. (B) Changein TCRb VJ SSS from donor to recipient at Day 100 and 1 year post-transplant. (C) RPD graphs representative of loss of self-similarity and order in recipient T cellrepertoire at day 100 following SCT, and modest recovery by 1 year post-SCT (compare with the donor data in Figure 1B for J and Figure 2 for VJ segment usage).Proportional distribution of the J segment and the VJ segments.

J. Meier et al. / Biol Blood Marrow Transplant 19 (2013) 366e377 375

proliferations [7]. However, none of these approachesprovides a comprehensive picture of T cell repertoirereconstitution in the recipient of SCT, knowledge ofwhich is critical to allow full comprehension of GVH andGVL responses, which are driven by minor histocompati-bility antigen (mHA) differences between donors andrecipients [23].

In this context, the observation of an alteration of clonalhierarchy in patients after SCT is particularly informative.After transplantation, a shift in clonal dominance wasobserved where by only a minority of TCR clones that weredominant in the infused stem cell product maintain thatposition in the new hierarchy, with a different set of clonesrising to high frequency. Although, sampling related incon-sistencies (donor apheresis product versus recipient bloodsample), the use of anti-thymocyte globulin during condi-tioning and post-SCT immunosuppression resulting indelayed T cell reconstitution most likely contributes to theobserved clone frequency variation, the substantial shift ofdominant clonotypes consistently seen within the top 4ranks in these HLA-matched patients suggests that newantigens (minor histocompatibility antigens) encountered inthe recipient may be responsible for driving this clonal shift.

The degree of disparity observed in patients without GVHD isintriguing, in that it suggests that along with other factors,anti-thymocyte globulin may facilitate tolerance induction to

Figure 7. (A) Model depicting fractal, self-similar TCR gene segment usage across the VDJ segments and concluding with NI in two dimensions. (B) Model portrayingTCR clonal ranking at different gene segment levels in the third dimension.

J. Meier et al. / Biol Blood Marrow Transplant 19 (2013) 366e377376

endogenous normal antigens in patients with minimalresidual disease, mitigating GVHDwithout impacting relapserates [24]. Although not evident in our dataset, given thelimited number of patients, such studies may provide insightinto the variable immunological sequela of SCT despitestringent HLA matching.

The thymus has a deterministic role to play in thisprocess, and decline in thymic function and T cell depletion,produced by immunosuppressive therapy after hematopoi-etic stem cell transplant in the presence of a host of newminor histocompatibility antigens, may help change thedominant T cell clones seen after transplantation. In theabsence of thymic selection (in older SCT recipients), thisrepertoire alterationmaywell contribute to the pathogenesisof complications such as GVHD. In fact, thymic selection maycontribute to the fractal organization of T cell repertoire bybalancing self-reactive T cells and regulatory T cells. It wasreported that naturally occurring regulatory T cells differ-entiate at the double positive stage in the thymus [24,25]. Inthe absence of the thymus, such modification of T cellrepertoire may be altered. This initial disordered organiza-tion in transplant recipients compared with that seen indonors may also reflect a more clonal nature of the TCRb reconstitution present in the early phases of transplantrecipient immune reconstitution. Whether the overall clonalhierarchy will be substantially different in patients who getpreparative regimens without T cell depletion using anti-thymocyte globulin remains to be determined. Accordingly,these results demonstrate a previously unexplainedcomplexity in the process of immune reconstitution at workin transplant recipients. Additionally, this finding also raisesthe intriguing possibility that infusions of selected donor-dominant TCR bearing clones of donor lymphocytes topatients with GVHD may be explored in the future on thebasis that these clones will be more likely to target antigensderived from pathogens rather than peptides of donor origin,and by extension, recipient. This discussion notwithstanding,our proposed model of the T cell repertoire organizationpertaining to the dominant clones remains hypothetical untilverified using alternative TCR Vb sequencing platforms.Recently evidence has emerged that there may be significantdifferences in how different sequencing platforms report thefinal repertoire diversity [26], and although these differencesare mostly observed in low-frequency clones and are cor-rected for in our data set, our conclusions should be

considered limited to the TCR sequencing methodology re-ported herein.

In conclusion, high-throughput sequencing of the T cellreceptor within the constraints of methodological andanalytic limitations, has unveiled a complex clonal frequencydistribution in the TCR repertoire of an individual. Logtransformation of T cell clone frequency demonstrates thatthe TCR b repertoire, as determined by TCR gene segmentusage is organized in a fractal, self-similar pattern witha hierarchy of dominant and minor clones. The complexity ofthe TCR repertoire is diminished after SCT, and that TCRb hierarchy is altered. Restoration of this order may serve asa therapeutic target to improve clinical outcomes by opti-mizing adoptive immunotherapeutic potential of allogeneicstem cell transplantation. Further, shifts in TCR clonaldominance may allow more accurate monitoring of immu-notherapy of malignancies in general, beyond allogeneicstem cell transplantation, in addition to monitoring ofimmunosuppressive therapies and responses to infections.Thus, studying the TCR repertoire, and accounting for itscomplexity, may yield new insights into the mechanism ofimmunological sequela observed in disease states such asthose seen after SCT.

ACKNOWLEDGMENTSThe authors gratefully acknowledge Drs. John McCarty,

Harold Chung and William Clark for their support in theconduct of the clinical trial these patients were enrolled on.

Financial disclosure: We acknowledge Genzyme Corpora-tion (Sanofi-Aventis) for the research funding to make thisstudy possible. Analytic services were provided by the VCUMassey Cancer Center Biostatistics Shared Resource, sup-ported in part with funding from NIH-NCI Cancer CenterSupport Grant P30 CA016059. A.T and. M.M. have receivedresearch support from Genzyme Corporation, a Sanoficompany.

Conflict of Interest Statement: Authors have no relevantconflicts of interest to disclose. M.M. and A.T. have receivedresearch funding from Genzyme-Sanofi. D.H., C.D. and C.S.are employed by Adaptive Biotechnologies.

Authorship Statement: M.M. and A.T. designed research.J.M., C.R., K.A., A.H., J.B., K.P., D.H., C.D., C.S., K.H., K.A., M.M.,and A.T. performed research. J.M. and A.T. analyzed data. J.M.,C.R., and A.T. wrote the report.

J. Meier et al. / Biol Blood Marrow Transplant 19 (2013) 366e377 377

SUPPLEMENTARY DATASupplementary data related to this article can be found at

http://dx.doi.org/10.1016/j.bbmt.2012.12.004.

REFERENCES1. West BJ. Physiology in fractal dimensions: error tolerance. Ann Biomed

Eng. 1990;18:135-149.2. Masters BR. Fractal analysis of the vascular tree in the human retina.

Annu Rev Biomed Eng. 2004;6:427-452.3. Glenny RW, Robertson HT, Yamashiro S, Bassingthwaighte JB. Appli-

cations of fractal analysis to physiology. J Appl Physiol. 1991;70:2351-2367.

4. Adam JA.Mathematics in Nature: Modeling Patterns in the Natural World.Princeton (NJ): Princeton University Press; 2003. p. 336-340.

5. Mandelbrot B. How long is the coast of Britain? Statistical self-similarity and fractional dimension. Science. 1967;156:636-638.

6. Toor AA, Sabo RT, Chung HM, et al. Favorable Outcomes in Patientswith High Donor-Derived T Cell Count after in Vivo T Cell-DepletedReduced-Intensity Allogeneic Stem Cell Transplantation. Biol BloodMarrow Transplant. 2012;18:794-804.

7. Berrie JL, Kmieciak M, Sabo RT, et al. Distinct oligoclonal T cells areassociated with graft versus host disease after stem-cell trans-plantation. Transplantation. 2012;93:949-957.

8. Robins HS, Campregher PV, Srivastava SK, et al. Comprehensiveassessment of T-cell receptor beta-chain diversity in alphabeta T Cells.Blood. 2009;114:4099-4107.

9. Robins H, Desmarais C, Matthis J, et al. Ultra-sensitive detection of rareT cell clones. J Immunol Methods. 2012;375:14-19.

10. West GB, Brown JH, Enquist BJ. The fourth dimension of life: fractalgeometry and allometric scaling of organisms. Science. 1999;284:1677-1679.

11. West GB, Brown JH. The origin of allometric scaling laws in biology fromgenomes to ecosystems: towards a quantitative unifying theory of bio-logical structure and organization. J Exp Biol. 2005;208:1575-1592.

12. Nevill AM, Bate S, Holder RL. Modeling physiological and anthropo-metric variables known to vary with body size and other confoundingvariables. Am J Phys Anthropol. 2005;(Suppl. 41):141-153.

13. Packard GC, Boardman TJ. Model selection and logarithmic trans-formation in allometric analysis. Physiol Biochem Zool. 2008;81:496-507.

14. Robins HS, Srivastava SK, Campregher PV, et al. Overlap and effectivesize of the human CD8þ T cell receptor repertoire. Sci Transl Med. 2010;2:47ra64. http://dx.doi.org/10.1126/scitranslmed.3001442.

15. Freeman JD, Warren RL, Webb JR, et al. Profiling the T-cell receptorbeta-chain repertoire by massively parallel sequencing. Genome Res.2009;19:1817-1824.

16. Naumov YN, Naumova EN, Hogan KT, et al. A fractal clonotype distri-bution in the CD8þ memory T cell repertoire could optimize potentialfor immune responses. J Immunol. 2003;170:3994-4001.

17. Naumova EN, Gorski J, Naumov YN. Simulation studies for a multistagedynamic process of immune memory response to influenza: experi-ment in silico. Ann Zool Fennici. 2008;45:369-384.

18. Jiang N, Weinstein JA, Penland L, et al. Determinism and stochasticityduring maturation of the zebrafish antibody repertoire. Proc Natl AcadSci USA. 2011;108:5348-5353.

19. Weinstein JA, Jiang N, White RA, et al. High-Throughput Sequencing ofthe Zebrafish Antibody Repertoire. Science. 2009;324:807-810.

20. Li H, Ye C, Ji G, Han J. Determinants of public T cell responses. CellResearch. 2012;22:33-42.

21. Raimondi G, Turner MS, Thomson AW, Morel PA. Naturally occurringregulatory T cells: recent insights in health and disease. Crit RevImmunol. 2007;27:61-95.

22. Bellucci R, Alyea EP, Weller E, et al. Immunologic effects of prophylacticdonor lymphocyte infusion after allogeneic marrow transplantation formultiple myeloma. Blood. 2002;99:4610-4617.

23. Shlomchik WD. Graft-versus-host disease. Nature Rev Immunol. May2007;7:340-352.

24. Portier DA, Sabo RT, Roberts CH, et al. Anti-thymocyte globulin forconditioning in matched unrelated donor hematopoietic cell trans-plantation provides comparable outcomes to matched related donorrecipients. Bone Marrow Transplant. 2012;47:1513-1519.

25. Nunes-Cabaço H, Caramalho I, Sepúlveda N, Sousa AE. Differentiationof human thymic regulatory T cells at the double positive stage. Eur JImmunol. 2011;41:3604-3614.

26. Bolotin DA, Mamedov IZ, Britanova OV, et al. Next generationsequencing for TCR repertoire profiling: Platform-specific features andcorrection algorithms. Eur J Immunol. 2012 Nov;42:3073-3083.